anu1meha/perfbench

perfbench

Welcome to the perfbench repository.

perfbench is an SDN controller/switch/hypervisor benchmarking tool.

Please pre-install hypervisors like FlowVisor, OpenVirteX, or OnVisor first as hypervisors.

It was tested with Ubuntu 14.04 up to 18.04.

Installation

Please run install.sh to completely install perfbench.

Compile

To compile perfbench go to the root directory and type cmake .;cmake --build . --target perfbench -- -j 4.

Run

Just run ./perfbench. Start with parameter --help to show command line parameters

For some testcases you need root privileges. Start perfbench with sudo ./perfbench

Here are some example use-cases. A working management network infrastructure between switch/hypervisor/perfbench is needed.

Switch Only

We recommend using OVS/Mininet as the switch. You can easily start a switch topology with:

sudo mn --topo=single,1 --controller=remote,ip=<IP_TO_PERFBENCH> --switch ovs,protocols=OpenFlow10

Then start perfbench it with the following parameters:

Here, we are using the FeaturesRequest testcase with a OpenFlow packet-rate of 1000 packets-per-second (pps) per tenant.

./perfbench --tenants <TENANT_COUNT> --testcase FeaturesRequest --pps 1000

Hypervisor

If you want to use perfbench with a hypervisor, start the switch first (OVS/Mininet):

sudo mn --topo=single,1 --controller=remote,ip=<IP_TO_PERFBENCH> --switch ovs,protocols=OpenFlow10

Then create the virtual SDNs (slicing, you can find it in the documentation of the hypervisor like FlowVisor)

Here we use a packet-rate of 100 pps with a send interval of 1000 milliseconds (ms). That means it sends bursts of 100 packets each second.

./perfbench --tenants <TENANT_COUNT> --testcase FeaturesRequest --pps 100 --send-interval 1000

Hypervisor with perfbench as Dataplane

We call this the 'Sandwich' use case.

If you want to use perfbench with a hypervisor and perfbench acting as the dataplane/switches, start it with the following parameters:

Non-Shared switches

Here we are using the FlowModSandwich testcase with an OpenFlow packet-rate of 100 pps per tenant per switch connection. We are using 5 switches per virtual-SDN/Tenant where all the tenants use their own switches (non-shared) (A setting with FlowVisor).

./perfbench --tenants <TENANT_COUNT> --testcase FlowModSandwich --num-switches 5 --pps 100 --switch-ip <IP_OF_HYPERVISOR>

Shared Switches

Here we are using the FlowModSandwich testcase with an OpenFlow packet-rate of 100 pps per tenant per switch connection. We are using 5 switches where all the tenants share the same switches.

./perfbench --tenants <TENANT_COUNT> --testcase FlowModSandwichSingle --num-switches 5 --pps 100 --switch-ip <IP_OF_HYPERVISOR>

Logging

Perfbench logs the transmitted/received (tx/rx) OpenFlow packets timestamps, type, XID and size in a file after each run. You can specify the logfile path with the: --logfile parameter. It automatically adds the tenant-id and "*.csv" to the end of the filename. e.g. --logfile path/to/dir/perfbench_log results in path/to/dir/perfbench_log_1.csv and path/to/dir/perfbench_log_2.csv if you used 2 tenants.

The logfile is structured as following (columns sepparated by " "):

xid	request_msgtype	request_timestamp	request_size	response_msgtype	response_timestamp	response_size	latency	conn_id
The xid of each OpenFlow Request/Response	Request Msgtype	Request Timestamp	Request Size (of OF packet)	Response Msgtype	Response Timestamp	Response Size	Latency in ms (Response Timestamp - Request Timestamp)	Connection-ID (Multi-Switch testcases)

e.g.

0 5 1444052493.642726 8 6 1444052493.643310 176 0.584126 0
1 5 1444052493.646452 8 6 1444052493.646632 176 0.180006 0
2 5 1444052493.655894 8 6 1444052493.656299 176 0.405073 0
...

Documentation

Please visit: https://wiki.lkn.ei.tum.de/public:perfbench:start for documentation